Bayesian Hierarchical Model for Analyzing Multiple Correlated Outcomes of Familial Data

• Activity Number: 664
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Qiaolin Chen*+ and Robert E Weiss and Catherine A Sugar
• Companies: University of California at Los Angeles and University of California at Los Angeles Fielding School of Public Health and University of California at Los Angeles
• Keywords: Bayesian Analysis ; hierarchical model ; Familial data ; Markov Chain Monte Carlo ; clustered outcomes ; schizophrenia

Abstract

We construct a Bayesian factor analysis model to analyze familial data of multiple correlated outcomes. Our primary focus is to estimate the difference in neurocognitive impairments between relatives of schizophrenia patients and relatives of community controls from the UCLA Family Neurocognitive Study (Asarnow 2002) which collected over 100 outcomes clustered in different domains. We build a factor analysis model including a shared factor for all family relationships and outcomes, specific factors for each family relationship, and specific factors for each outcome which are assumed to be independent from each other. Family relationship specific factors and outcome specific factors are allowed to be correlated among themselves. The patterned factor loading matrix is the same as the multitrait multimethod matrix. Next, we put different constraints and priors on the factor loading matrix to get a full Bayesian model and different intermediate models. Likelihood and posterior probabilities are calculated and Markov Chain Monte Carlo method is used for computation. These models are then combined using Bayesian model averaging.